AI spend
Go to E-learning → AI spend.
The AI tutor, the practice-paper generator and the voice features all cost money every time a learner uses them. This page is where you see how much, and it warns you before a month gets away from you.
You need the finance admin or AI admin right to open it.
What the page tells you
Section titled “What the page tells you”The banner at the top says whether you are within budget, approaching it, or over. It compares today against a daily budget and this month against a monthly one.
Spend on
On track for is the important one during a busy revision period. It takes what has been spent so far this month and projects the whole month at that rate, so you can act while there is still a month left to change. Next to it, last 7 days is the figure to trust when the month started quietly — a slow first week drags the projection down for a fortnight.
What drove it breaks the day down by feature, biggest first. Generating a practice paper is by far the most expensive thing a learner can do; a chat message is a tiny fraction of it.
Who was spending splits the same money between learners on a free trial and paying learners. This is the number to watch during a free-trial campaign, because trial spend has no revenue behind it.
Paper reuse cache shows how often a learner got a paper that had already been generated for somebody else. Those cost nothing. The percentage matters because papers are shared across every learner: if the way papers are built ever changes, the cache empties and the next learner on each kind of paper pays for a fresh generation. This number tells you what that change would cost.
Alarms
Section titled “Alarms”You should not have to open this page to find out you are over budget.
When spend crosses a threshold, an alarm goes out by SMS to the people on the operations list, by email where email is set up, and into the server log. It fires once — a problem that is still true tomorrow is not sent again every few minutes — but it always fires again if a warning becomes serious.
The Alarms panel lists what has fired recently and, in the right-hand column, which channels actually accepted each one. If it says reached nobody, the alarm was recorded but not delivered, and somebody needs to add a recipient.
Send a test alarm (AI admins only) fires a message that says it is a test, and tells you which channels accepted it. Use it when you set the page up, and again if the recipients change. It sends a real SMS, so do not press it repeatedly.
Reading the figures honestly
Section titled “Reading the figures honestly”The panel headed How to read these figures is not boilerplate — read it once.
- Measured means the AI provider counted the usage itself and told us. Trust these.
- Estimated means we worked it out. Voice calls are always an estimate: the audio goes straight from the learner’s browser to the provider, so nothing on our side ever learns how long a call lasted. For voice, the number of sessions is the fact; the money is a model.
- Free is a cache hit — a real request that cost nothing.
Prices are the AI vendors’ published list prices converted at a fixed exchange rate, not figures from an invoice. Treat the totals as the right order of magnitude with a reliable trend, and reconcile against the real bill once a month.
One thing the page deliberately does not do is count requests and call that spend. One generated exam paper uses roughly fifty times what a chat message uses, yet each is a single request — so a request count would be badly wrong in the direction that makes things look cheap.
If the page says AI spend could not be read, that is a connection problem, not a quiet day. The figures are unavailable, not zero.
Asking Waridi instead
Section titled “Asking Waridi instead”The same numbers are available in plain language. Ask the assistant:
- What did AI cost us yesterday?
- Are we on track for this month’s AI budget?
- Is the exam paper cache working?
Waridi reads the same meter and will tell you which figures are measured and which are estimated.